Fimeg/letta-server
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

fix: patch gemini token counting (#6445)

Browse Source 
fix: use usage_metadata.candidates_token_count for counting total tokens
This commit is contained in:
Charles Packer
2025-11-28 18:52:21 -08:00
committed by Caren Thomas
parent c6df306ccf
commit e142d440d5
1 changed files with 5 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
letta/interfaces/gemini_streaming_interface.py
View File

@@ -166,8 +166,11 @@ class SimpleGeminiStreamingInterface:
if usage_metadata:
if usage_metadata.prompt_token_count:
self.input_tokens = usage_metadata.prompt_token_count
if usage_metadata.total_token_count:
self.output_tokens = usage_metadata.total_token_count - usage_metadata.prompt_token_count
# Use candidates_token_count directly for output tokens.
# Do NOT use (total_token_count - prompt_token_count) as that incorrectly
# includes thinking/reasoning tokens which can be 10-100x the actual output.
if usage_metadata.candidates_token_count:
self.output_tokens = usage_metadata.candidates_token_count
if not event.candidates or len(event.candidates) == 0:
return